Thread (127 messages) 127 messages, 14 authors, 2022-12-01

Re: [PATCH 38/44] KVM: Disable CPU hotplug during hardware enabling

From: Huang, Kai <hidden>
Date: 2022-11-10 01:33:38
Also in: kvm, kvm-riscv, kvmarm, linux-mips, linux-riscv, linux-s390, linuxppc-dev, lkml

On Wed, 2022-11-02 at 23:19 +0000, Sean Christopherson wrote:
quoted hunk ↗ jump to hunk
From: Chao Gao <redacted>

Disable CPU hotplug during hardware_enable_all() to prevent the corner
case where if the following sequence occurs:

  1. A hotplugged CPU marks itself online in cpu_online_mask
  2. The hotplugged CPU enables interrupt before invoking KVM's ONLINE
     callback
  3  hardware_enable_all() is invoked on another CPU right

the hotplugged CPU will be included in on_each_cpu() and thus get sent
through hardware_enable_nolock() before kvm_online_cpu() is called.

        start_secondary { ...
                set_cpu_online(smp_processor_id(), true); <- 1
                ...
                local_irq_enable();  <- 2
                ...
                cpu_startup_entry(CPUHP_AP_ONLINE_IDLE); <- 3
        }

KVM currently fudges around this race by keeping track of which CPUs have
done hardware enabling (see commit 1b6c016818a5 "KVM: Keep track of which
cpus have virtualization enabled"), but that's an inefficient, convoluted,
and hacky solution.

Signed-off-by: Chao Gao <redacted>
[sean: split to separate patch, write changelog]
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/x86.c  |  8 +++++++-
 virt/kvm/kvm_main.c | 10 ++++++++++
 2 files changed, 17 insertions(+), 1 deletion(-)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index a7b1d916ecb2..a15e54ba0471 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -9283,7 +9283,13 @@ static int kvm_x86_check_processor_compatibility(struct kvm_x86_init_ops *ops)
 	int cpu = smp_processor_id();
 	struct cpuinfo_x86 *c = &cpu_data(cpu);
 
-	WARN_ON(!irqs_disabled());
+	/*
+	 * Compatibility checks are done when loading KVM and when enabling
+	 * hardware, e.g. during CPU hotplug, to ensure all online CPUs are
+	 * compatible, i.e. KVM should never perform a compatibility check on
+	 * an offline CPU.
+	 */
+	WARN_ON(!irqs_disabled() && cpu_active(cpu));
 
Also, the logic of:

	!irqs_disabled() && cpu_active(cpu)

is quite weird.

The original "WARN(!irqs_disabled())" is reasonable because in STARTING section
the IRQ is indeed disabled.

But this doesn't make sense anymore after we move to ONLINE section, in which
IRQ has already been enabled (see start_secondary()).  IIUC the WARN_ON()
doesn't get exploded is purely because there's an additional cpu_active(cpu)
check.

So, a more reasonable check should be something like:

	WARN_ON(irqs_disabled() || cpu_active(cpu) || !cpu_online(cpu));

Or we can simply do:

	WARN_ON(!cpu_online(cpu) || cpu_active(cpu));

(because I don't know whether it's possible IRQ can somehow get disabled in
ONLINE section).

Btw above is purely based on code analysis, but I haven't done any test.
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help